Company / organization
Qwen
By and about Qwen
Anthropic Research: A global workspace in language models
Anthropic open-sourced the Jacobian lens, an interpretability tool that reads what Claude is about to say from its internal layers before output, finding a brain-like "global workspace" divide in the model's computation.
newsDSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation
DeepSeek open-sourced DSpark, a confidence-scheduled speculative decoding method that delivers 51% to 400%+ throughput boosts on its V4 models and works on Gemma and Qwen too.
newsAlibaba: Qwen 3.7
Alibaba unveiled Qwen3.7-Max with 1M-token context, claiming benchmark wins over Claude Opus 4.6, while multimodal Qwen3.7-Plus launched at $0.40/$1.60 per million tokens alongside custom Zhenwu M890 chips.
newsGoogle: Gemma 4 12B and DiffusionGemma
Google released Gemma 4 (12B to 31B, Apache 2.0, natively multimodal, 256K context) for local deployment, and followed with DiffusionGemma, a 26B diffusion model generating 1,000+ tokens/sec on a single H100 by denoising in parallel instead of…
newsUpcoming Events: AI Socratic Europe & China Chapter Tour
Fed and Anissa are launching AI Socratic chapters across Europe and China, meeting with local leaders in London, Lisbon, and Berlin before spending a month in China visiting Alibaba, Qwen, x.AI, and robotics factories.
newsThinking to Recall: How Reasoning Unlocks Parametric Knowledge in LLMs
Google Research finds that enabling reasoning in LLMs helps them recall simple facts through two mechanisms: extra tokens provide computational buffer time, and intermediate facts prime the correct answer—but hallucinated intermediate facts sharply…
newsAlibaba Qwen 3.5 Small Model Series
Alibaba released four Qwen 3.5 small models ranging from 0.8B to 9B parameters with native multimodal capabilities, then saw its lead researcher Junyang Lin and three others unexpectedly resign a day later.
newsStepFun's Step 3.5 Flash
StepFun released Step 3.5 Flash, a sparse mixture-of-experts model with 196B total parameters but only 11B activated per token, designed to run on 128GB of memory and trained with the Muon optimizer.
newsDo LLMs Benefit From their own Words?
MIT researchers found that LLMs degrade in long conversations by treating their own previous responses as fact, causing errors to compound, and that removing prior AI outputs often improves quality while cutting context length by up to 10x.
newsQwen3 Coder Next
Alibaba's Qwen released Qwen3 Coder Next, an open-weight language model designed for coding agents and local development workflows.
newsRecursive Language Models: LLMs that call themselves to handle arbitrarily long prompts
Researchers introduce Recursive Language Models, which let LLMs decompose and recursively call themselves to process inputs up to 100 times longer than their native context windows, outperforming GPT-5 by 26% on long-context tasks.
newsUS AI startups increasingly built on Chinese open-source foundations
Chinese open-source models like DeepSeek and Qwen now lead US models in global downloads with 17% market share versus 15.8%, attracting US startups with superior price-to-performance but raising concerns about embedded censorship and regulatory risk.
newsSimWorld: an open-ended simulator for agents in physical and social worlds
Researchers created SimWorld, a simulator where AI models compete in a market economy with tasks like food delivery; Claude and Qwen used riskier strategies with higher returns while other models played conservatively, and Qwen and DeepSeek undercut…
newsQwen3 Coder 30B A3B Instruct
Alibaba released Qwen3 Coder 30B A3B Instruct on July 31, 2025, a 30.5B parameter mixture-of-experts model with 256K token context designed for code generation, repository understanding, and tool use.
newsCornell's vec2vec translates embeddings between models it has never seen
Cornell researchers' vec2vec translates text embeddings between models with no paired data or encoder access, hitting up to 0.92 cosine similarity and 100% top-1 accuracy, and recovering readable info from up to 80% of test emails via zero-shot inversion.
newsManus: DeepResearch + Operator agent from China
Chinese startup Butterfly Effect released Manus, an autonomous AI agent powered by Claude 3.5 and Qwen that performs real-world tasks like building websites and analyzing stocks, outperforming OpenAI's Deep Research on the GAIA benchmark.